Structural representation of pronunciation and its use in pronunciation training

1 Introduction No two students are the same. Pronunciation teaching should be started after teachers know exactly how individual students are in their development. In one-to-fifty situations such as classrooms, it is difficult for a teacher to know their pronunciations precisely. Computers have provided us with tools to visualize the pronunciations based on acoustics. In this case, however, good knowledge of acoustics is needed. Further, the acoustic representation inevitably shows many things irrelevant to the proficiency, such as speaker individuality, gender, age, microphone differences and so on. The first author already proposed a unique method of representing speech acoustics, where dimensions of the non-linguistic factors are well removed, Minematsu (2004a). This representation can be regarded as physical implementation of structural phonology, where only the interrelations among speech sounds are focused. In this paper, the non-native pronunciations are described based on the new method and the development of a student's pronunciation is traced. Further, automatic assessment of the pronunciation is investigated experimentally.